Wireless Sensor Network intrusion detection based on Borderline-SMOTE and Deep Ensemble Learning

With the development of the Internet of Things (IoT), Wireless Sensor Networks (WSNs) are widely used in several fields. Due to limited node resources and vulnerability to attacks, AI-driven intrusion detection systems have become the main means. However, the imbalance of data categories and the diversity of attack types lead to limited detection performance, so for the problems of scarce attack samples in the dataset and insufficient performance and generalization ability of traditional models, this paper proposes a WSN intrusion detection method that combines Borderline-SMOTE oversampling and deep ensemble learning. The method effectively balances the sample distribution and ensemble LSTM, CNN and ANN models to significantly improve the detection accuracy and robustness. Experimental results show that based on the WSN-DS dataset, the method achieves an accuracy of 99.20% in detecting multiple attack types, which is better than the traditional methods, especially in the F1 scores for Blackhole, Grayhole, and Flooding attacks, providing an effective solution for WSN security.